Confusion over WORKING_DIRECTORY

[icolwell-as]

Hi There, I recently spent a lot of time trying to understand why some of my unit tests that worked fine in ROS2 Dashing were failing in ROS2 Foxy. I hope the below helps other people who may come across the same issue.

According to Dashing, Foxy, and Humble documentation, the following is stated about the default WORKING_DIRECTORY for the ament_add_gtest() macro.

The default working directory otherwise is the CMAKE_SOURCE_DIR, which will be evaluated to the directory of the top-level CMakeLists.txt.

However, this only seems to be true for Dashing (maybe the intermediate releases too, I didn't use them). PR #206 changed the default behaviour from CMAKE_SOURCE_DIR to CMAKE_CURRENT_BINARY_DIR.

This leaves me with a few questions:

1. What is the best way to use the old behaviour? In my case, I have test data/files in the source repo but that data isn't available in CMAKE_CURRENT_BINARY_DIR. I also have about 10-15 ament_add_gtest() calls. Is there some way to set the default once instead of adding WORKING_DIRECTORY "${CMAKE_SOURCE_DIR}" to every ament_add_gtest() call? Or is it better practice to copy test data into the build directory only, but not the install directory?
2. Is there a better source of documentation for ament/ament_cmake? or is docs.ros.org the best place to be looking?

I guess there aren't any action items for this repo, but I suppose I could open a documentation ticket that references this one.
I think Foxy was the first release to have this guide? I'm not sure.

[clalancette]

However, this only seems to be true for Dashing (maybe the intermediate releases too, I didn't use them). PR #206 changed the default behaviour from CMAKE_SOURCE_DIR to CMAKE_CURRENT_BINARY_DIR.

Yes, very good call.

1. What is the best way to use the old behaviour? In my case, I have test data/files in the source repo but that data isn't available in CMAKE_CURRENT_BINARY_DIR. I also have about 10-15 ament_add_gtest() calls. Is there some way to set the default once instead of adding WORKING_DIRECTORY "${CMAKE_SOURCE_DIR}" to every ament_add_gtest() call? Or is it better practice to copy test data into the build directory only, but not the install directory?

I think our current practice is to just specify it for every test. I don't know that we have a way to set it once. See https://github.com/ros2/rcl/blob/rolling/rcl_yaml_param_parser/CMakeLists.txt , for example.

• Is there a better source of documentation for ament/ament_cmake? or is docs.ros.org the best place to be looking?

docs.ros.org is the right place. If you would open a pull request to fix up https://github.com/ros2/ros2_documentation/blob/rolling/source/How-To-Guides/Ament-CMake-Documentation.rst , I'd be happy to review it.

[icolwell-as]

Thank you very much @clalancette for the fast response!

I'll open a ticket in the documentation repo and get a PR up within the next couple of days.

The example you provided is now making me wonder if we should revert #206? I understand it's nice to follow cmake behaviour, but if the majority of the ROS community is running tests with data/files, maybe CMAKE_SOURCE_DIR is a more appropriate default behaviour?

[clalancette]

The example you provided is now making me wonder if we should revert #206? I understand it's nice to follow cmake behaviour, but if the majority of the ROS community is running tests with data/files, maybe CMAKE_SOURCE_DIR is a more appropriate default behaviour?

I will say that at least in the core, we only need to set WORKING_DIRECTORY in a few places; most places don't need it. Looking through all of the packages released into Rolling, it looks like only ~22% need to use it[1]. Additionally, I don't think it is a good idea to switch back and forth all of the time; that just causes churn in the ecosystem.

Given all of that, my preference would be to leave things as-is and fix up the documentation.

[1] I calculated this number by pulling all of the sources of packages released into Rolling, counting the number of ament_add_gtest instances (558), counting the number of WORKING_DIRECTORY instances that used SOURCE (126), and then dividing. This won't be an exact number; some of the instances of ament_add_gtest are going to be the actual definition of it, so the actual number is lower. Also, I'm not sure that all of these tests pass. Still, I think this is good enough for an estimate.

[icolwell-as]

Thanks @clalancette, I agree, probably best to leave it as-is for now.
It can always be re-addressed in the future if there truly comes a need for it.

Thanks for your thoughts! I'll close this out and try to update the documentation soon.